Introduction
CH4 and N2O emissions together comprise 38 % of the total global
radiative forcing attributable to emissions of well-mixed greenhouse gases
(Myhre et al., 2013). N2O is also a major component of stratospheric chemical cycles, acting as the
largest contributing species towards stratospheric ozone depletion, and
predicted to remain so throughout the 21st century (Ravishankara et al., 2007). CH4 emissions can
also lead to the formation of tropospheric ozone through reaction with OH
radicals, leading to air quality issues associated with potentially
dangerous respiratory problems in many cities across the world (Ebi and McGregor, 2008).
The globally averaged atmospheric abundances of CH4 and N2O have
increased respectively from 722±25 to 1803±2 ppb and 270±7 to 324.2±0.2 ppb in the period 1750–2011 (Hartmann et al., 2013). However, the relative
contribution of individual sources and sinks to the atmospheric abundance of
both species is highly uncertain (Ciais et
al., 2013; Kirschke et al., 2013). Top-down, atmospheric measurement-based
approaches can provide important constraints on these global budgets, both
through direct estimation of sectorially and/or regionally disaggregated
emissions using atmospheric inversion models (Fraser
et al., 2013; Thompson et al., 2014) and by enabling validation of the
process models used to compile bottom-up emission inventories (Krinner
et al., 2005; O'Shea et al., 2014b). Representative sampling on regional and
national scales can also act as an important aid to establishing effective
emission reduction policies at both national and international levels.
In situ aircraft-based measurements form an important part of this top-down
approach, enabling high-resolution sampling on regional scales
(e.g. O'Shea et al., 2013a), vertical profile measurement (e.g. Wofsy et
al., 2011), and sampling in remote regions far from ground stations (e.g. Kort et al., 2012).
Greenhouse gas flux estimates can then be made using mass balance
(Karion et al., 2013; O'Shea et al., 2014a; Peischl et al., 2015),
eddy covariance (Ritter et al., 1992; Hiller et al., 2014; Yuan et al., 2015) or inverse modelling techniques
(Kort et al., 2008; Polson et al., 2011; Xiang et al., 2013), the latter frequently in
association with ground-based measurements (Miller et al., 2013). Aircraft
measurements can also be used to validate both ground-based and
satellite-based remote sensing techniques, forming an important link across
a wide range of spatial and temporal measurement scales (Tanaka
et al., 2012; Wecht et al., 2012). However, it should be noted that of the
studies listed above, only Wofsy et al. (2011) and Xiang et al. (2013) made
continuous in situ measurements of N2O, emphasising the need for wider
deployment of in situ instrumentation to measure N2O on aircraft.
During summer 2014, the FAAM (Facility for Airborne Atmospheric
Measurements) large atmospheric research aircraft (hereafter referred to as
the FAAM aircraft) participated in the GAUGE (greenhouse gas UK and global
emissions) and MAMM (methane and other greenhouse gases in the Arctic:
measurements, process studies and modelling) measurement campaigns. This
aircraft component of the GAUGE campaign focussed on greenhouse gas
measurement around the UK to allow emission estimates to be made in
conjunction with both inverse modelling and mass balance techniques. An
important element of this campaign was to better constrain emissions from
the agricultural sector, which is the second largest contributor (after the
energy sector) towards the UK's total greenhouse gas emissions, producing
N2O through the use of nitrogen-based fertilisers and CH4 by
enteric fermentation in livestock (Webb et al., 2014). The MAMM
campaign focussed on improving understanding of Arctic CH4 emissions,
dominated by biogenic emission from natural wetlands (Zhuang et al., 2006), in order to
help better constrain measurement-derived global CH4 budgets and to
allow comparison against the emissions predicted by regional land surface
models (O'Shea et al., 2014b). Accurate measurement of CH4 and N2O on board the FAAM
aircraft was therefore of critical importance during these campaigns.
Infrared (IR) spectroscopy is frequently employed for airborne measurement
of greenhouse gas mole fractions (Chen
et al., 2010; O'Shea et al., 2013b; Santoni et al., 2014), enabling high
frequency measurement (usually ≥1 Hz) and fast instrument response
times (of the order seconds). Many instruments make use of the superior
lasers, optics and detectors available in the near-IR region around
∼6000 cm-1 (Baer et al., 2002; Crosson,
2008). However, line strengths for CH4 and N2O are of the order
100 and 100 000 times stronger respectively in the mid-IR spectral region of
∼1000–4000 cm-1 (Rothman
et al., 2013). For CH4 these competing effects lead instruments
operating in both spectral ranges to achieve broadly comparable
performances. For N2O, however, the comparatively weak line strengths
in the near-IR, coupled with the lower atmospheric abundance of N2O,
make mid-IR spectroscopy much more suitable for atmospheric measurement.
Rannik et al. (2015) find that the best long-term and short-term precisions for N2O measurement are obtained
using continuous-wave quantum cascade laser (QCL)-based instruments, which
operate in the mid-IR region.
In this paper we discuss the development of an airborne measurement system
for CH4 and N2O, using a mid-IR, continuous-wave, quantum cascade laser absorption spectrometer (QCLAS, Aerodyne Research Inc., USA) on board
the FAAM aircraft. We focus on measurements from the GAUGE and MAMM
campaigns conducted during summer 2014.
Details of the direct absorption spectroscopy and associated spectral
retrieval algorithm employed are given in Sect. 2, including the empirically
derived correction for the presence of water vapour. The configuration and
optimisation of the sample and calibration air flow systems for airborne
measurement are also presented in this section. In Sect. 3, calibration
procedures used to tie the data to the WMO (World Meteorological Office)
greenhouse gas scale are described and assessed, both through analysis of
in-flight calibration data and comparison with simultaneous CH4
measurements using a Fast Greenhouse Gas Analyser (FGGA, model RMT-200, Los
Gatos Research, USA; described by O'Shea et al.,
2013b). Section 4 presents N2O data from GAUGE flight B868 in more
detail and outlines how these data could be used in combination with a mass
balance technique to estimate a regional N2O flux for the north-west of
England. Finally, the key findings of this work are summarised in Sect. 5.
Operational design
Instrument specification
In this section we briefly describe the operating principles of the QCLAS
used to measure N2O and CH4 on board the FAAM aircraft. This
instrument uses a mid-IR, thermoelectrically cooled, continuous-wave,
distributed feedback QCL (Alpes Laser, Switzerland) as a light source. The
laser beam is directed through an astigmatic Herriott multipass absorption
cell, providing an effective optical pathlength through the sample of 76 m
(McManus et al., 1995), and collected by a thermoelectrically cooled
photovoltaic detector (Vigo Systems, Poland). The
output frequency of the QCL is scanned over a small spectral region
(1275.3–1275.8 cm-1), containing rotational–vibrational N2O, CH4 and
H2O line transitions, by repeatedly ramping the laser current whilst
holding the laser at a constant temperature. The laser is swept across this
region at a rate of ∼5 kHz, with a measurement of the
detector's zero-light output (noise-equivalent signal) recorded at the end
of each sweep by dropping the current below the laser threshold (such that
there is no emission from the laser). Because the linear current ramp does
not produce a precisely linear frequency response from the laser, it is
necessary to determine the tuning rate using a Germanium etalon, which can
be placed in the path of the beam before the multipass cell.
Schematic showing the QCLAS air sampling and data handling systems.
The C.E.P.E. (calibration, exhaust, power and electronics) unit and the
Aerodyne QCL mini-monitor enclosures are represented by dashed boxes around
the components they contain. The calibration cylinders are labelled “low”,
“high” and “target”. The flow rate of the calibration gas is controlled by
the MFC (mass flow controller), and the flow rate into the instrument is
monitored by the MFM (mass flow meter). The optical components associated
with the alignment of the laser beam are not shown.
The laser temperature is held at ∼-23 ∘C
using a Peltier cooler. Excess heat is removed using a coolant fluid, which
is recirculated and maintained at ∼25 ∘C by a
CustomChill thermoelectric liquid chiller (CRAL300DHP-12, CustomChill, USA).
The optical layout of the instrument is described in detail by Nelson
et al. (2004), although our use here of a continuous-wave laser rather than
a pulsed laser allows for removal of the beam splitter and the incorporation
of two additional mirrors to aid the adjustment of beam alignment prior to
entering the cell.
TDLWintel software
The laser current control and mole fraction retrievals were performed using
the TDLWintel software package, details of which are provided by
Nelson et al. (2002). In brief, this retrieval
relies on the Beer–Lambert law, given by
I(ν)I0(ν)=exp(-α(ν,P,T)nCl),
where l is the path length of the beam through the absorber, C is the
concentration of sample gas, n is the absorber mole fraction and α(ν,P,T) is the frequency-, pressure- and temperature-dependent absorption cross section of the absorber. The intensity I(ν)
is measured by the detector, which also measures the background intensity
I0(ν) at window wavelengths outside the wings of target absorption
lines. A polynomial fit is applied to the data obtained at these
non-absorbing wavelengths such that variation in baseline intensity
measurement, due to changes in both laser output and detector sensitivity
across the measured spectral range, can be subtracted from the spectrum
(Zahniser et al., 1995).
In order to determine the mole fraction of a target species at a sampling
rate of 1 Hz, TDLWintel fits a Voigt line shape profile to an averaged
spectrum, consisting of ∼5000 individual laser sweeps, using
a Levenberg–Marquardt retrieval algorithm. Line strengths and positions are
taken from the HITRAN 2012 database (Rothman
et al., 2013). The pressure and temperature of the sample are continuously
measured by in situ sensors positioned within the sample gas flow on the
outlet of the cell, allowing air broadening effects to be considered in the
retrieval.
Configuration for airborne measurement
The QCLAS is mounted on a rack inside the cabin of the FAAM aircraft, with a
rearward-facing, 3/8′′ outer diameter, stainless steel inlet inserted through
a customised window blank (Avalon Aero Ltd., UK). The sample flow line
consists of Swagelok 1/4′′ outer diameter PFA Teflon tubing, partly encased
within the inlet, and forming a pressure seal via a bored-through Swagelok
3/8′′ to 1/4′′ reducing union. Figure 1 shows a schematic of the QCLAS air
sampling system including the configuration for delivering calibration gas
to the sample cell. The sample flow line is ∼2.5 m in length
from the inlet tip to the pressure controller. Sample flow rate is measured
using a 30 SLPM (standard litres per minute) mass flow meter
(M10MB01334CS3BV, MKS Instruments UK Ltd, UK), placed directly upstream of a
0.5 µm sintered particle filter (SS-4F-05, Swagelok, USA).
The choice of sample cell pressure is a balance between two effects: higher
pressures increase the absorption, thus improving the signal-to-noise ratio
of the measurement, whilst pressure broadening of the spectral lines
increases spectral overlap and line mixing (as discussed by
Zahniser et al., 1995). Airborne operation is also complicated by the large
variation in inlet pressures typically encountered over the course of a
flight (down to ∼250 hPa at 10 km altitude).
Control of the cell pressure is provided by an electronic pressure
controller (640A-13TS1V62V, MKS Instruments UK Ltd, UK) placed upstream of
the sample cell, as shown in Fig. 1. This maintains a constant pressure by
automatically adjusting an internal valve to restrict the flow of air
through it. As the inlet pressure decreases, the valve is set to a
progressively more open position. The minimum inlet pressure that can be
sampled whilst maintaining any given cell pressure is attained when the
pressure controller valve reaches its fully open position. This minimum
inlet pressure is then equal to the sum of the (chosen) cell pressure and
the pressure drops across each component of the inlet system (including the
fully open pressure controller valve). Hence choosing a lower cell pressure
decreases this minimum inlet pressure, enabling the cell to be held at
constant pressure up to a higher altitude. A cell pressure of 68.9±3.6 hPa (at 1σ) was used during the GAUGE and MAMM campaigns.
Air is pulled through the system using a single stage scroll pump (Edwards
XDS10, Edwards, UK). A throttle valve (253B-1-40-1, MKS Instruments UK Ltd,
UK) positioned between the sample cell outlet and the pump inlet is used to
control the flow rate through the system. This again is a balance between
the desire to decrease the instrument response time, favouring a faster flow
rate, and the desire to reduce the total pressure drop between the inlet and
the sample cell, favouring a slower flow rate. Throughout the GAUGE and MAMM
campaigns the valve was set to 18 % of its fully open position, resulting
in a constant mass flow rate of 1.43±0.21 SLPM (at 1σ) down
to inlet pressures of ∼380 hPa. At lower inlet pressures both
the mass flow rate and the cell pressure were reduced.
Laboratory tests were performed to establish the effect of cell pressure
changes on the mole fractions retrieved when sampling a compressed air
cylinder. The variability in retrieved mole fraction was found to be no
greater across cell pressures typically encountered during high altitude
flying (inlet pressures below 380 hPa; cell pressures down to
∼46 hPa) than across the range experienced during low
altitude flying (inlet pressures above 380 hPa; cell pressures between 65
and 76 hPa). It was therefore deemed unnecessary to filter data
according to the absolute cell pressure value. However, rapid changes in
pressure were found to impact upon the retrieved mole fractions;
consequently data were removed whenever the 10 s standard deviation in cell
pressure exceeded 0.1 Torr (13.3 Pa).
The response time of the system was determined in the laboratory (at 1017 hPa
inlet pressure) by overflowing the inlet with N2 from a compressed
gas cylinder. The e-folding time of the system was determined using an
exponential fit to the decay in retrieved mole fractions and was found to
be 0.68±0.12 s (mean ±1σ). The inlet lag time was
given by the time between turning on the N2 flow and the first drop in
retrieved mole fractions. In this laboratory test it was found to be in the
range 2–3 s; however, we expect this lag time to decrease with
altitude (up to ∼380 hPa) as the volumetric flow rate of the
system scales inversely with air density (for a constant mass flow rate).
In principle the Beer–Lambert relationship described above can be used to
retrieve absolute mole fractions. However, in-flight calibration is commonly
used to account for instrumental drift when using optical-based measurements (e.g.
O'Shea et al., 2013b; Santoni et al., 2014), as external variables such as
temperature and pressure can undergo significant variation during a flight.
Our system employs three calibration standards to scale the data and assess
instrument performance, as described in Sect. 3.1.
The three calibration standards are stored in 10 L carbon fibre hoop-wrap
cylinders (BFC 124-136-002, Aluminium Alloy 7060, Luxfer, UK), which are
filled to ∼300 bar and mounted to the QCLAS rack. Each
cylinder is fitted with a high-pressure brass valve (C215, Rotarex,
Luxembourg), screwed into the cylinder collar using PTFE
(polytetrafluoroethylene) tape. An all brass adapter connects this to
instrument-grade stainless steel tubing of 1/8′′ outer diameter, specially
cleaned for high purity service (21512, Thames-Restek, UK). This tube joins
each cylinder to a separate single-stage diaphragm brass regulator
(44-2212-244-1382, TESCOM, UK). On the low-pressure side of the regulator,
Swagelok 1/8′′ outer diameter PFA Teflon tubing is used. Three three-way valves
(009-0294-900, Parker-Hannifin, USA) and one two-way valve (009-0089-900,
Parker-Hannifin, USA) are used to select the flow from the desired
calibration cylinder (a fourth port allows sampling from an external
cylinder). The flow rate is set using a mass flow controller
(1179A01314CS1BV, MKS Instruments UK Ltd, UK) to provide an overflow of
calibrant at the inlet (upstream of the mass flow meter).
Water vapour correction
The influence of water vapour on spectral retrievals can be very significant
(e.g. Allen et al., 2014),
particularly given the wide range of natural water vapour concentrations
typically encountered over the course of a flight (from a small fraction of
a percent to many percent in the troposphere alone). To ensure comparability
between measurements made at different humidity levels it is necessary to
remove this effect and report dry mole fractions.
Many opt to circumvent the need to correct for this influence by drying the
sample air before it enters the instrument, often using a combination of
Nafion gas dryers and dry ice traps (e.g.
Daube et al., 2002; Peischl et al., 2010; Santoni et al., 2014). The
advantage of this approach is obvious, as any empirically derived correction
for the influence of water vapour will contribute, often significantly, to
the overall uncertainty of the measurements. However, there are several
disadvantages associated with drying the sample, as discussed in detail by
Rella et al. (2013). Of particular
relevance for the QCLAS system described here are the issues associated with
increasing the pressure drop across the inlet system, increasing the
residence time in the inlet system and the logistical problems of supplying
dry ice to remote field locations and transporting it in a sealed cabin
environment.
In cases such as this, where the sample is not dried, an empirical
correction must be derived in order to account for the water vapour
influence. Typically this involves applying a scale factor to the retrieved
mole fractions, with its form and coefficients determined through laboratory
experiment. Rella et al. (2013), O'Shea et al. (2013b) and Zellweger et al. (2012) employ this approach across a
variety of spectroscopic instruments. However, a recently added feature of
the TDLWintel software allows the effect of line broadening (in addition to
sample dilution) due to the presence of water vapour to be included in the
spectroscopic retrieval itself. Instrument-specific coefficients quantifying
the broadening due to water vapour must be derived empirically for each
species. These coefficients are equal to the dimensionless ratio of the line
broadening due to water vapour pressure to the line broadening due to dry
air pressure. Water-corrected mole fractions are then determined by first
retrieving the water vapour mole fraction, then combining this result with
the appropriate water broadening coefficients in the retrieval of the other
species.
Here we compare the effectiveness of these two approaches in the case of the
QCLAS. Such a comparison of these two methods is instructive to other
experimentalists that may seek to apply similar corrections. Both approaches
require empirical (laboratory) data, which can be obtained simultaneously
for direct comparison.
As the empirical coefficients required by both methods can be determined
using the same experiment, and TDLWintel retains the measured raw
spectral data, it was possible to perform this comparison by reanalysing the
same data set, either deriving a scale factor to post-process the data or
varying the way in which the water vapour influence was treated in the
retrieval. The data here were gathered in four separate laboratory
experiments, each using an identical experimental setup to that employed by
O'Shea et al. (2013b), who provide a full
description. In summary, dry compressed air was humidified to a variety of
different water vapour mole fractions, spanning the range of 0–2 %
typically measured in flight. Between each measurement of humid air, a dry
reference was sampled, using a dry ice trap to reduce the sample dew point
to less than -60 ∘C. It should be noted that by re-drying the
air downstream of the humidifier we accounted for any dissolution of gases
in the humidifier and the temperature dependence of this effect.
The first approach used the following relationship to scale the measurements
of the wet sample to corresponding dry mole fractions
Xdry=Xweta+bH2O,
where Xwet and Xdry are the raw and the scaled water-corrected
mole fractions respectively for species X, and H2O represents the
retrieved mole fraction of water vapour. Coefficients a and b were
derived by performing a weighted least orthogonal distance regression of
Xwet/Xdry as a function of H2O for data from all four
experiments.
The empirical values for the QCLAS were found to be a=1.00096,
b=-0.0154563 %-1 for N2O and a=1.00071,
b=-0.0136929 %-1 for CH4. The uncertainties associated with these values can be
quantified by the residuals of this regression for CH4 and N2O,
shown in Fig. 2. The RMS (root mean square) values for these residuals are
2.5 and 0.50 ppb for CH4 and N2O respectively.
It is apparent from Fig. 2 that substantially different behaviour was
observed on 20 June 2014 when compared to the three other experiments. This
is likely to be associated with a lack of long-term stability in the
retrieval of H2O mole fraction, as indicated by the drift in the
average retrieved H2O mole fraction during the dry runs seen in Fig. 2.
As this is a measurement of very dry air (less than -60 ∘C dew
point), this drift can be assumed to represent the variability in the
baseline intensity in the region of the H2O absorption line, likely
resulting from very small changes in the optical alignment and/or
pathlength. This subtle variability in baseline intensity then manifests as
systematic variability in the accuracy of the retrieved H2O mole
fractions. The consequence of this on the long-term stability of the scale
factor method, and the improvement gained using the spectroscopic
correction, is discussed below.
Residual error due to the influence of water vapour for the
retrieval of (a) CH4 and (b) N2O after applying an empirically
derived scale factor to correct the data. Data from four identical
experiments are shown; the residuals are calculated as the product of the
fractional error for each measurement and the average mole fraction for the
dry measurements taken during all four experiments.
The second, spectroscopic, water vapour correction method used the water
broadening function in TDLWintel to correct for the influence of water
vapour. Reanalyses of the raw spectra were performed using a variety of
different water broadening coefficients in the retrieval. For each
coefficient, the difference between the retrieved wet mole fraction and the
corresponding dry measurement was calculated at every water vapour level
used during the four experiments. The RMS difference for each coefficient,
averaged over the entire data set, is shown in Fig. 3. It can be seen that
the correction performed best using water broadening coefficients of 1.6 and
1.8 for CH4 and N2O respectively. These optimal coefficients
resulted in RMS differences between corresponding wet and dry measurements
of 1.6 ppb for CH4 and 0.32 ppb for N2O; these are the values used
to determine the contribution of uncertainty in this water vapour correction
to the total measurement uncertainty of the instrument.
It was thus concluded that in this case a better correction for the
influence of water vapour was obtained using the spectroscopic correction
performed by the TDLWintel software than was achieved by scaling the wet
mole fractions according to Eq. (2). There are two potential factors that
could explain the improved performance of the spectroscopic method over the
scale factor method. Firstly, because the spectroscopic correction
determines the water vapour pressure broadening relative to the dry air
pressure broadening, it implicitly accounts for changes in absolute water
vapour pressure associated with corresponding changes in measured sample
cell pressure. In contrast, the scale factor method using Eq. (2) relies
only on the retrieved mole fraction of water vapour and so fails to account
for any increase or decrease in water vapour pressure broadening at higher
or lower cell pressures.
RMS (root mean square) difference between the retrieved wet mole
fractions and the corresponding dry measurements for CH4 and N2O,
as a function of water broadening coefficient. These RMS values are
determined using data taken over the full experimental range of H2O
mole fraction during all four identical experiments.
Secondly, in the scale factor method, drift in the uncalibrated water vapour
mole fraction measurements propagates directly via Eq. (2) into a systematic
error in the water-corrected mole fraction for both CH4 and N2O
(Xdry). In the spectroscopic method, inaccuracies in the measurement of
water vapour mole fraction instead impact upon the subsequent retrieval of
CH4 and N2O by affecting the fitting of the Voigt line profile.
Although inaccurately calculated water vapour line broadening will change
the retrieved mole fraction using the spectroscopic method, because the
water vapour broadening is just one of several parameters constraining the
spectral fit, inaccuracy resulting from this effect will be manifest in part
as a reduction in the goodness of fit (spectral residual). Hence the
spectroscopic method is less sensitive to any potential drift in the
retrieval of water vapour mole fraction than the scale factor method.
All flight data presented in this paper have been reanalysed using this
spectroscopic water vapour correction, with empirically derived water
broadening coefficients of 1.6 and 1.8 for CH4 and N2O
respectively.
Data quality
Systematic instrumental error associated with changes in external variables
such as temperature and pressure can be compensated for by repeated sampling
of calibration gas. During airborne sampling an instrument is exposed to
rapid changes in these variables over a wide range of values; hence regular
calibration is required.
In this section we first describe the calibration procedure used during the
two campaigns and explain the rationale behind it. We then seek to diagnose
and understand the sources of systematic error which remain uncaptured by
this calibration. Finally, we describe an alternative calibration procedure
designed to better address these key sources of error and evaluate the
effect of both methods on the overall data quality.
Original calibration procedure
The in-flight calibration procedure employed throughout the GAUGE and MAMM
campaigns was in principle similar to that described by O'Shea et al. (2013b). The data were scaled using
two cylinders of known composition, traceable to the WMO greenhouse gas
scale (WMO, 2009), whose mole fractions spanned the normal
measurement range for N2O and CH4. By sequentially pumping gas
from these cylinders through the system and comparing the retrieved mole
fractions to their WMO-traceable values, two reference points could be
established for the QCLAS on the WMO scale. By assuming a linear
relationship, the “true” mole fraction corresponding to each retrieved QCLAS
mole fraction was given by interpolating the scale between the two reference
points. For each calibration a scale factor (Mx) and zero offset
(Cx) were found using
Mx=Xhigh,WMO-Xlow,WMOXhigh,meas-Xlow,meas,Cx=Xhigh,WMO-MxXhigh,meas,
where Xhigh,WMO and Xlow,WMO are the “true”
WMO-traceable mole fraction values, and Xhigh,meas and
Xlow,meas are the measured mole fraction values, for the high
and low calibration cylinders respectively for a given species X.
These two cylinders were sampled sequentially on an approximately hourly
basis and the values for Mx and Cx were linearly interpolated
between calibrations. The raw data were then calibrated by applying
Xcal(t)=Mx(t)Xraw(t)+Cx(t).
In order to check that both interpolation between the two cylinder mole
fraction values and temporal interpolation between hourly calibrations were
justified, a third WMO-traceable “target” cylinder containing intermediate
CH4 and N2O mole fractions was measured approximately mid-way
between the hourly high–low span calibrations. Applying the above
calibration to this target cylinder measurement and comparing the resulting
calibrated mole fractions with the WMO-traceable values for the cylinder
enabled errors associated with this method to be quantified. Raw CH4
data demonstrating a typical calibration cycle are shown in Fig. 4.
A selection of raw CH4 data from flight B848, overlaid with
calibration index markers to highlight the hourly calibration cycle. The
target cylinder measurement (green markers) is performed approximately
mid-way between the high–low span cylinder measurements (black and blue
markers respectively) used to calibrate the data to the WMO scale.
The offset between the raw QCLAS CH4 data and the calibrated
FGGA data (used here as our reference) during flight B848. Gradients of over
30 ppb in less than 10 min can be seen to be present.
This calibration procedure was designed to remove linear drifts acting over
timescales of the order of the inter-calibration time, here approximately
1 h. However, analysis of the difference in CH4 mole fraction between
the raw QCLAS data and the calibrated data from the on-board FGGA frequently
showed gradients of over 30 ppb in timescales of less than 10 min, as
shown for flight B848 in Fig. 5. The FGGA on board the FAAM aircraft has
previously been shown not to exhibit any significant systematic errors on
this timescale (O'Shea et al., 2013b), suggesting
that these gradients represent a source of systematic error in the QCLAS
data. Note that although we have compensated here for the lag time between
the two instruments using the correlation between the two CH4 data sets,
large deviations from the overall trend with very short durations are
present as a result of small differences in the measurement time of large
CH4 enhancements.
The offset between the raw QCLAS CH4 measurements and both the
corresponding calibrated FGGA data and the known contents of the target,
high and low calibration cylinders during flight B848, shown as a function
of static pressure. Although the absolute magnitude of the offset differs
for these four different measurements, the same broadly repeatable pattern
is exhibited by each of them.
Figure 6 shows the same CH4 data from flight B848 plotted as a function
of static pressure, as measured by the aircraft's RVSM (reduced vertical
separation minimum) system. It can be seen that a broadly repeatable pattern
exists as a function of pressure, which was found to dominate variability in
the raw QCLAS CH4 data offset with respect to the FGGA over the course
of the flight. The same pattern is also exhibited by the offset of the
calibration cylinder measurements from the nominal values of the cylinders
(also shown), although the absolute value of this offset clearly differs
between the cylinders. Similar patterns were observed for the cylinder
measurements of N2O, and across other flights during the two campaigns.
The fact that this variation with pressure can be observed in both the raw
sample data and the measurement of dry calibration air confirms that errors
in the water vapour correction cannot be responsible. A leak (ingress) into
the system also appears implausible, as one would expect this to have the
opposite effect on the high and low cylinder measurements, pulling both
towards the mole fractions of CH4 and N2O present in the aircraft
cabin. Santoni et al. (2014) warn of issues associated with fluctuations in sample cell
pressure. However, the offsets in retrieved QCLAS mole fractions observed
here were found to exhibit no dependence on either sample cell pressure,
sample cell temperature or the rate of change for these variables (as
recorded by the pressure and temperature sensors within the sample
flow).
The temperature of the cabin air was also recorded as it entered the outer
enclosure of the QCLAS, but this again exhibited no clear correlation with
the CH4 data offset. The pressure inside the cabin, however, was not
recorded during the 2014 campaigns. Subsequent test flights, described in
Sect. 3.2 below, suggest that it was variability in this quantity that
caused the large gradients in CH4 offset described above.
The offset between the raw retrieved QCLAS mole fractions and the
known content of a target cylinder, sampled continuously during three
separate deep profiles during flight B903. Panels (a) and (b) shown the offset
as a function of the pressure inside the cabin, for CH4 and N2O
respectively. Panels (c) and (d) show the offset as a function of external
static pressure (also for CH4 and N2O respectively).
Influence of cabin pressure
In April 2015 we performed a test flight (B903) designed to further
understanding of the underlying issues behind the large gradients in QCLAS
CH4 data described in Sect. 3.1 above. This time cabin pressure data
were available throughout the flight, and three deep profiles were performed
whilst the QCLAS sampled compressed air from a calibration cylinder. The
first deep profile occurred at the beginning of the flight, whilst the other
two were performed ∼2.5 h later.
Figure 7 shows the offset in retrieved CH4 and N2O mole fractions
from the known composition of the cylinder as a function of both cabin
pressure and RVSM external static pressure. For external pressures between
∼800 and ∼1000 hPa these offsets remain
roughly constant; this can be seen to correspond to the range of external
pressures over which there is no change in cabin pressure. The fact that,
for both CH4 and N2O, the raw QCLAS measurement offset does not
change when the cabin pressure is constant, even when the external pressure
is varying, strongly suggests that cabin pressure variability is the primary
cause of the large gradients in this offset observed throughout the GAUGE
and MAMM campaigns in summer 2014.
Time series from flight B903, showing the offset between the raw
retrieved QCLAS mole fractions and the known content of the target cylinder
being sampled. Cabin pressure and external static pressure are also shown to
illustrate the systematic nature of the offset during two consecutive
profiles.
A time series of the retrieved mole fraction offsets for the final two
profiles is shown in Fig. 8. The influence of changing cabin pressure on
these retrieved mole fractions can be characterised as a small-scale
oscillation superimposed on a larger-scale gradient. This large-scale
gradient appears to be very consistent across the three profiles (see also
Fig. 7), whereas the small-scale oscillations are not so repeatable.
The likely pathway through which cabin pressure can influence the retrieved
mole fractions is through its effect on baseline intensity in the spectral
regions close to the absorption lines. There are two potential components to
this: the effect of changing absorption in the open path section of the
laser and the effect of changing pressure on the alignment of, and spacing
between, the instrument's optical components.
To investigate the effect of varying the open path absorption, a further
test flight was conducted whilst flowing dry nitrogen through the laser beam
enclosure. A simulation was also performed, where open path CH4 and
N2O absorption were included in the spectral fit for the range of cabin
pressures encountered during flight B903, to assess the impact on the
retrieved mole fractions. Both of these tests indicated that varying open
path absorption contributed negligibly to the observed gradients in
retrieved mole fraction over the range of cabin pressures experienced during
these flights. Hence we conclude that these gradients are more likely
attributable to small changes in optical alignment/spacing associated with
cabin pressure variation.
Pressure-differentiated calibration procedure
As the short-term (of order minutes) instrumental drift with pressure had a
greater effect in degrading measurement precision than any longer-term (of
order hours) drift with time, the data were reanalysed using an alternative
calibration procedure, designed to reduce the impact of this issue on the
overall accuracy of the calibrated measurements. As there were no cabin
pressure data available for the GAUGE and MAMM campaigns in summer 2014, it
was necessary to use the external static pressure from the aircraft's RVSM
system as a proxy. This approach is justified by the strong correlation
between cabin pressure and external pressure and by the results in Sect. 3.4 below.
In this approach values of Mx(t) and Cx(t) for sections of flight
at broadly equivalent pressure levels (defined here as a range of
variability less than 15 hPa for a period longer than 2 min) were
interpolated between any calibrations conducted at a pressure within 15 hPa
of the average pressure during that section. Profile data, along with all
data at pressure levels where no calibrations were performed, were flagged
as poor quality and removed from the analysis. This pressure-differentiated
calibration method has the disadvantages of both reducing the amount of
calibrated data for the campaigns by 54 % and potentially inducing errors
associated with long-term instrumental drift, as data can be separated from
the corresponding calibration(s) by up to 5 h. The effect on the overall
data quality of using this pressure-differentiated calibration procedure is
discussed and compared in Sect. 3.4.
Mean and standard deviation of the difference between QCLAS 1 Hz
target cylinder measurements and the nominal cylinder values and the
difference between the 1 Hz QCLAS and the corresponding 1 Hz FGGA sample
CH4 measurements, using both the original and pressure-differentiated
calibration methods.
QCLAS–target difference (ppb)
QCLAS–FGGA
difference (ppb)
N2O
CH4
CH4
Calibration method
Mean
1σ
Mean
1σ
Mean
1σ
Original
0.00319
0.960
0.253
4.78
-2.87
8.27
Pressure-differentiated
0.105
0.419
0.0668
1.71
-2.05
5.85
Histograms showing the offset between the calibrated 1 Hz QCLAS
measurements and both the corresponding target cylinder values and the
corresponding FGGA sample measurements. Panels (a)–(c) show histograms for
data calibrated using the original method; in comparison, panels
(d)–(f) show the corresponding histograms using the pressure-differentiated
calibration method. It can be seen here that the pressure-differentiated
method results in the removal of many of the outlying target cylinder
measurements. In addition, the Gaussian fit to the QCLAS–FGGA CH4
offset is also improved by the pressure-differentiated calibration method
(f) relative to the original method (c).
It was also found that large roll angles (∼20∘ or
over), associated with sharp turns of the aircraft, produced short-term
deviations in retrieved CH4 and N2O mole fractions, evident in
both the raw and calibrated data. It is likely that this effect is a
consequence of slight alignment changes (similar to Sect. 3.2 above) caused
by the centrifugal force of the turn (no relationship with cabin pressure
variability was found). Whilst this effect was clearly secondary to the
pressure-dependent variability described above, producing CH4
deviations of less than 5 ppb, it was decided to flag all data associated
with roll angles of greater than 10∘ as reduced quality. All
calibrated data (using both methods) discussed here have been filtered
according to this flag, removing the reduced-quality data associated with
high roll angles. The application of this filter reduces the total size of
the raw data set by only 7 %.
Results and discussion
The performance of the QCLAS can be assessed both by comparing the
calibrated target cylinder measurements to their corresponding WMO-traceable
values and by comparing the calibrated 1 Hz CH4 sample data with the
corresponding measurements from the on-board FGGA. No other instruments on
board the FAAM aircraft measured N2O during the GAUGE or MAMM
campaigns, so a direct comparison of sample N2O mole fractions cannot
be made here. Table 1 summarises these results for both the original
calibration method described in Sect. 3.1 and the pressure-differentiated
calibration method described in Sect. 3.2.
It can be seen from the table that using the pressure-differentiated
calibration method significantly improves the accuracy of the QCLAS, both
during target cylinder measurements and sample mode. In particular, the
standard deviations in QCLAS–target and QCLAS–FGGA differences are
substantially reduced compared to the equivalent values produced using the
original calibration method. The WMO recommends compatibility between
different analyses within 2 ppb for CH4 and 0.1 ppb for N2O
(WMO, 2013). The fraction of data within these ranges for both
the QCLAS–target and QCLAS–FGGA differences using both methods is shown in
Table 2. Here again it can be seen that the pressure-differentiated
calibration method produces superior results.
The fraction of 1 Hz data within the WMO compatibility
recommendations for QCLAS target cylinder measurement and QCLAS–FGGA sample
measurement, using both the original and pressure-differentiated calibration
methods.
QCLAS–target
QCLAS–FGGA
Calibration method
N2O
CH4
CH4
Original
0.149
0.519
0.292
Pressure-differentiated
0.309
0.765
0.361
Allan precision for QCLAS measurement of CH4 and N2O,
both during ambient in-flight sampling and whilst sampling a compressed air
cylinder in the laboratory, given for averaging times of 1, 10 and 108 s.
1σ Allan precision (ppb)
1 s
10 s
108 s
Flight
Lab
Flight
Lab
Flight
Lab
CH4
0.52
0.48
0.31
0.17
0.23
0.12
N2O
0.11
0.12
0.074
0.044
0.042
0.029
Figure 9 shows the offset between the calibrated 1 Hz QCLAS target cylinder
measurements and the known content of the cylinder, as histograms for both
CH4 and N2O. It can be seen here that the improved standard
deviations obtained using the pressure-differentiated calibration method
result from the removal of outlying data associated with the pressure effect
discussed in the previous section. Also shown are histograms of the
QCLAS–FGGA offset for 1 Hz CH4 sample data, which provide further
evidence of the superior performance of the pressure-differentiated
calibration method. The data produced using the original calibration method
are clearly far less well represented by a Gaussian fit; this is to be
expected in the presence of a systematic effect such as that described in
Sect. 3 above. In contrast, the Gaussian shape of the
pressure-differentiated data is consistent with a random error distribution
for both instruments.
The instrument precision can be quantified using the Allan variance
technique (Werle et al., 1993). Table 3 presents the 1σ Allan precision (over 1, 10 and 108 s) for
CH4 and N2O, both in a laboratory environment whilst sampling a
compressed air cylinder and in flight during a period of ambient background
sampling. These results are similar to those of Santoni et al. (2014),
with in-flight 1 Hz precisions here of 0.52 ppb for CH4 and
0.11 ppb for N2O.
Finally, a nominal uncertainty for the data can be calculated using the
known uncertainties from the water vapour correction experiment, the
calibration of the target cylinder to the WMO-scale and the in-flight target
measurements. Table 4 contains these values for both CH4 and N2O
using the pressure-differentiated calibration method. The nominal total
uncertainties for CH4 and N2O are ±2.47 and ±0.54 ppb respectively.
Known component and nominal total uncertainties for the QCLAS
measurement of CH4 and N2O, calibrated using the
pressure-differentiated method.
1σ uncertainty (ppb)
Water vapour
Target standard
In-flight target
Total
correction
calibration
measurements
CH4
1.63
0.77
1.71
2.47
N2O
0.32
0.11
0.42
0.54
Aircraft flight track for flight B868, coloured by N2O mole
fraction. Average wind speeds and directions taken over 60 s are shown as a
wind barbs (using the convention where each full barb represents a wind
speed of 10 knots). Selected HYSPLIT back trajectories are shown for a
region of enhanced N2O (black) and a contrasting region of lower
N2O (grey). Map data: Google, SIO, NOAA, US Navy, NGA, GEBCO,
Landsat.
Case study
The GAUGE project aims to provide top-down greenhouse gas emission estimates
for the UK, which can be used to validate the bottom-up inventory-based
estimates required by UK and international legislation. As part of this
approach, the use of aircraft data is planned in combination with mass
balance techniques (Karion et al., 2013; O'Shea et al., 2014a;
Peischl et al., 2015) to estimate regional greenhouse
gas emissions. Such analysis is beyond the scope of this technical study;
however, we present QCLAS data from a single flight here as exemplars of
typical flight data, providing context with regard to scientific case
studies which may use this new airborne data set.
Flight B868 was designed to incorporate upwind and downwind sampling over
north-western England to provide a data set for a mass balance case study. This
region contains several large urban areas (Manchester, Liverpool, Leeds and
Sheffield) and also includes several areas of agricultural activity, known
to be an important anthropogenic source of N2O (Syakila
and Kroeze, 2011). The flight track, coloured by N2O mole fraction, is
shown in Fig. 10. The wind speed and direction, as measured on board the
aircraft and taken as an average over 60 s, are represented by wind barbs,
according to the standard convention where each full barb represents a wind
speed of 10 knots. Selected Lagrangian back trajectories using the HYSPLIT
(Hybrid Single-Particle Lagrangian Integrated Trajectory) model
(Draxler and Hess, 1998) are also shown (representing
around 24 h track across the UK mainland in the figure). These
trajectories were initiated using endpoints and trajectory end times
selected along the flight track and modelled with full vertical dynamics
using Global Data Assimilation System 1∘ resolution data.
It can be seen that the N2O mole fractions measured in the north-west of
the domain were enhanced relative to those in the south-east. A relatively
consistent south-easterly wind direction (both measured and seen in the
trajectories) suggests that this enhancement may enable the use of a mass
balance technique to estimate the N2O flux from an area between this
downwind transect and the corresponding upwind measurements using the
techniques described by O'Shea et al. (2014a). This
requires suitable investigation of the necessary assumptions, which is
beyond the scope of this simple example.
It is also striking that there is a relative contrast in the south-west area
of the domain, with N2O mole fractions around 328 ppb (compared with
330 ppb in the north-west). The potential reasons for this small contrast, in
terms of air-mass history, may be explained by considering both the
trajectories and the wind barbs. The trajectories from the north-west domain
show recent transport at low altitude (below 1 km) over Greater Manchester
and the north-west conurbation, whereas south-west trajectories pass over
more rural areas. This appears counterintuitive, as we expect the
agricultural sector to be the primary contributor towards N2O emissions
in this region. However, the wind barbs (which represent real measurements)
show that there is a complex divergence in the wind field in the south-west
domain, perhaps indicative of a localised sea-breeze circulation that cannot
be expected to have been captured at the resolution of the meteorological
data that were used to initialise the HYSPLIT trajectories. This sea-breeze
circulation could suggest recirculation of maritime air and hence dilution
of any moderately enhanced air arriving on the prevailing wind from the
east. The differing localised dynamics and air-mass histories of the two
domains may explain the observed contrast. Further analysis of this may form
the basis of future work and this limited example demonstrates the utility
of aircraft data in understanding local and regional air-mass
characteristics.